Run Keras models in the browser, with GPU support using WebGL

Last update: Dec 29, 2022

Overview

**This project is no longer active. Please check out TensorFlow.js.**
The Keras.js demos still work but is no longer updated.

Run Keras models in the browser, with GPU support using WebGL

Interactive Demos | Documentation

Run Keras models in the browser, with GPU support provided by WebGL 2. Models can be run in Node.js as well, but only in CPU mode. Because Keras abstracts away a number of frameworks as backends, the models can be trained in any backend, including TensorFlow, CNTK, etc.

Library version compatibility: Keras 2.1.2

Interactive Demos

Check out the demos/ directory for real examples running Keras.js in VueJS.

Basic Convnet for MNIST
Convolutional Variational Autoencoder, trained on MNIST
Auxiliary Classifier Generative Adversarial Networks (AC-GAN) on MNIST
50-layer Residual Network, trained on ImageNet
Inception v3, trained on ImageNet
DenseNet-121, trained on ImageNet
SqueezeNet v1.1, trained on ImageNet
Bidirectional LSTM for IMDB sentiment classification

Documentation

MIT License

Comments

Error: [Model] error loading weights.

I'm running the latest Keras JS on a trained Keras model. Following the instructions given on the README, I tested loading my model before training (random initialized weights) and it works.

However, after I perform training and use the exact same way to called model.save_weights, Keras JS is unable to read the model file. When trying to parse the metadata, it gives [Model] error loading weights.

Any insight into why this would happen?

opened by calclavia 10
Multi-dimensional Sequential Inputs / unflattened float32 arrays

This library appears to be focused on the Model structure, which is okay, but most of my trained networks are in Sequential form. Specifically LSTM Sequentials.

LSTM Sequentials generally expect a three-dimensional input (batchSize, X,Y). However this repo requires there to be only one flattened input for Sequential, named "input". Has anyone found a good way to convert Sequentials to a Model or a way around this?

I would like to just do prediction in javascript. That means the model shouldn't even need to be compiled, so maybe I can just hack something together myself.

opened by LRonHubs 9

Error loading demo model ("RangeError: Invalid typed array length")

Hi, I am trying to load the mnist_vae model weights demo with the following html:

<html>
<head>
  <meta charset="utf-8">
  <script language="javascript" type="text/javascript" src="scripts/lib/keras.js"></script>
</head>
<body>
  <script language="javascript" type="text/javascript">
	  const model = new KerasJS.Model({
	  filepaths: {
	    model: 'static/model/mnist_vae.json',
	    weights: 'static/model/mnist_vae_weights.buf',
	    metadata: 'static/model/mnist_vae_metadata.json'
	  },
	  gpu: false
	});
  </script>
</body>
</html>

This is the error I get in Chrome:

RangeError: Invalid typed array length
    at new Float32Array (native)
    at http://0.0.0.0:8000/scripts/lib/keras.js:4:18505
    at Array.map (native)
    at http://0.0.0.0:8000/scripts/lib/keras.js:4:18250
    at Array.forEach (native)
    at t.value (http://0.0.0.0:8000/scripts/lib/keras.js:4:16687)
    at http://0.0.0.0:8000/scripts/lib/keras.js:4:15092
From previous event:
    at t.value (http://0.0.0.0:8000/scripts/lib/keras.js:4:15067)
    at new t (http://0.0.0.0:8000/scripts/lib/keras.js:4:14730)
    at http://0.0.0.0:8000/keras.html:8:18

I get a similar error in Firefox that begins with:

Error: WebGL: Exceeded 16 live WebGL contexts for this principal, losing the least recently used one.
RangeError: attempting to construct out-of-bounds TypedArray on ArrayBuffer
Stack trace:
[...]

Any ideas what might be going wrong?

opened by mobeets 7

Weights have been permuted for Convolution1D

I noticed an error building Convolutional1D layer in js when the input has 1 feature layer. I added a failing test for it to this fork, so that it can be easily reproduced.

Here is the error reported at localhost:3000/test (all other tests pass)

Error: cwise: Arrays do not all have the same shape!
    at assign_cwise_thunk (eval at createThunk (eval at <anonymous> (http://localhost:3000/dist/keras.js:2858:1)), <anonymous>:8:71)
    at Object.assign_ndarrayops [as assign] (eval at makeOp (eval at <anonymous> (http://localhost:3000/dist/keras.js:108:1)), <anonymous>:3:43)
    at Convolution2D._w2row (eval at <anonymous> (http://localhost:3000/dist/keras.js:493:1), <anonymous>:274:30)
    at Convolution2D.setWeights (eval at <anonymous> (http://localhost:3000/dist/keras.js:493:1), <anonymous>:126:12)
    at Convolution1D.setWeights (eval at <anonymous> (http://localhost:3000/dist/keras.js:1355:1), <anonymous>:103:20)
    at r.<anonymous> (convolutional/Convolution1D.js:61:19)
    at r (https://cdnjs.cloudflare.com/ajax/libs/mocha/3.0.2/mocha.min.js:2:7852)
    at r.run (https://cdnjs.cloudflare.com/ajax/libs/mocha/3.0.2/mocha.min.js:2:8853)
    at i.runTest (https://cdnjs.cloudflare.com/ajax/libs/mocha/3.0.2/mocha.min.js:2:13553)
    at https://cdnjs.cloudflare.com/ajax/libs/mocha/3.0.2/mocha.min.js:2:14192
    at r (https://cdnjs.cloudflare.com/ajax/libs/mocha/3.0.2/mocha.min.js:2:13024)
    at https://cdnjs.cloudflare.com/ajax/libs/mocha/3.0.2/mocha.min.js:2:13000
    at n (https://cdnjs.cloudflare.com/ajax/libs/mocha/3.0.2/mocha.min.js:2:12791)
    at https://cdnjs.cloudflare.com/ajax/libs/mocha/3.0.2/mocha.min.js:2:12855
    at i (https://cdnjs.cloudflare.com/ajax/libs/mocha/3.0.2/mocha.min.js:1:559)

opened by wassname 7

Conv1D layer: check for legacy shape of weights

Keras changed the ordering of the weights (see https://github.com/fchollet/keras/commit/9f6acd960c0a0c699c79ca1d571783e1692568fb and https://github.com/fchollet/keras/commit/305b3bed747bb8dd358cf82d11bcb1aee5b6e517).

This will only transpose the weights if in legacy format. Fixes issue #22.

opened by stekaiser 6
Fira sans is broken in Chrome and Firefox on Linux

On Debian testing at least but probably on other distros as well. Could be also broken on Mac, the issue with Fira sans was revealed in habrahabr.ru redesign (worked fine on Windows but was totally invisible on Mac/Linux). It looks like this: Disabling font-family in body reveals the text. Please replace the font with the working version.

opened by rkfg 4
Demo is not working

webgl.js:56Uncaught TypeError: Cannot read property 'getShaderPrecisionFormat' of null keras.js:4 Uncaught TypeError: Cannot read property 'getParameter' of null bundle.js:15n @ bundle.js:1(anonymous function) @ bundle.js:1(anonymous function) @ bundle.js:1

opened by AlexanderKozhevin 4
Can not initiate model in nodejs with demo models
Hi

I am trying to predict on the IMDB bidrectional LSTM demo in NodeJS, so I have this code:

const KerasJS = require('keras-js'); const model = new KerasJS.Model({ filepaths: { model: 'imdb_bidirectional_lstm.json', weights: 'imdb_bidirectional_lstm_weights.buf', metadata: 'imdb_bidirectional_lstm_metadata.json' }, filesystem: true });

When I run this, it throws this error:

RangeError: Invalid typed array length at Float32Array (native) at D:\development\webserving\node_modules\core-js\modules_typed-array.js:416:15 at new Float32Array (D:\development\webserving\node_modules\core-js\modules\es6.typed.float32-array.js:3:12) at D:\development\webserving\node_modules\keras-js\lib\Model.js:465:30 at Array.map (native) at D:\development\webserving\node_modules\keras-js\lib\Model.js:433:37 at Array.forEach (native) at Model._createLayers (D:\development\webserving\node_modules\keras-js\lib\Model.js:352:19) at D:\development\webserving\node_modules\keras-js\lib\Model.js:195:15 at tryCatcher (D:\development\webserving\node_modules\bluebird\js\release\util.js:16:23) at Promise._settlePromiseFromHandler (D:\development\webserving\node_modules\bluebird\js\release\promise.js:512:31) at Promise._settlePromise (D:\development\webserving\node_modules\bluebird\js\release\promise.js:569:18) at Promise._settlePromise0 (D:\development\webserving\node_modules\bluebird\js\release\promise.js:614:10) at Promise._settlePromises (D:\development\webserving\node_modules\bluebird\js\release\promise.js:693:18) at Promise._fulfill (D:\development\webserving\node_modules\bluebird\js\release\promise.js:638:18) at PromiseArray._resolve (D:\development\webserving\node_modules\bluebird\js\release\promise_array.js:126:19)

Any idea what I am doing wrong here?

Thanks :-)
opened by spec2e 3

simply importing keras.js results in "Uncaught SyntaxError: Invalid regular expression"

Hi transcranial, I just wanted to write a simple web page to import keras-js. After cloning this repo, I create a file test.html containing the following codes:

<html>
<head>
<title>keras-js test</title>

<script src="dist/keras.js"></script>

<script type="text/javascript">

function periodic() {
  var d = document.getElementById('egdiv');
  d.innerHTML = 'Random number: ' + Math.random()
}

var net; // declared outside -> global variable in window scope
function start() {

  setInterval(periodic, 500);
}

}
</script>
</head>

<body onload="start()">

<div id="egdiv"></div>

</body>
</html>

Then I dragged this file into my Chrome browser, but in the console I received the following error:

Uncaught SyntaxError: Invalid regular expression: /[ÂªÂµÂºÃ€-Ã–Ã˜-Ã¶Ã¸-ËË†-Ë‘Ë -Ë¤Ë¬Ë®Í°-Í´Í¶Í·Íº-Í½Î†Îˆ-ÎŠÎŒÎŽ-Î¡Î£-ÏµÏ·-ÒÒŠ-Ô§Ô±-Õ–Õ™Õ¡-Ö‡×-×ª×°-×²Ø -ÙŠÙ®Ù¯Ù±-Û“Û•Û¥Û¦Û®Û¯Ûº-Û¼Û¿ÜÜ’-Ü¯Ý-Þ¥Þ±ßŠ-ßªß´ßµßºà €-à •à šà ¤à ¨à¡€-à¡˜à¢ à¢¢-à¢¬à¤„-à¤¹à¤½à¥à¥˜-à¥¡à¥±-à¥·à¥¹-à¥¿à¦…-à¦Œà¦à¦à¦“-à¦¨à¦ª-à¦°à¦²à¦¶-à¦¹à¦½à§Žà§œà§à§Ÿ-à§¡à§°à§±à¨…-à¨Šà¨à¨à¨“-à¨¨à¨ª-à¨°à¨²à¨³à¨µà¨¶à¨¸à¨¹à©™-à©œà©žà©²-à©´àª…-àªàª-àª‘àª“-àª¨àªª-àª°àª²àª³àªµ-àª¹àª½à«à« à«¡à¬…-à¬Œà¬à¬à¬“-à¬¨à¬ª-à¬°à¬²à¬³à¬µ-à¬¹à¬½àœààŸ-à¡à±à®ƒà®…-à®Šà®Ž-à®à®’-à®•à®™à®šà®œà®žà®Ÿà®£à®¤à®¨-à®ªà®®-à®¹à¯à°…-à°Œà°Ž-à°à°’-à°¨à°ª-à°³à°µ-à°¹à°½à±˜à±™à± à±¡à²…-à²Œà²Ž-à²à²’-à²¨à²ª-à²³à²µ-à²¹à²½à³žà³ à³¡à³±à³²à´…-à´Œà´Ž-à´à´’-à´ºà´½àµŽàµ àµ¡àµº-àµ¿à¶…-à¶–à¶š-à¶±à¶³-à¶»à¶½à·€-à·†à¸-à¸°à¸²à¸³à¹€-à¹†àºàº‚àº„àº‡àºˆàºŠàºàº”-àº—àº™-àºŸàº¡-àº£àº¥àº§àºªàº«àº-àº°àº²àº³àº½à»€-à»„à»†à»œ-à»Ÿà¼€à½€-à½‡à½‰-à½¬à¾ˆ-à¾Œá€€-á€ªá€¿á-á•áš-áá¡á¥á¦á®-á°áµ-á‚á‚Žá‚ -áƒ…áƒ‡áƒáƒ-áƒºáƒ¼-á‰ˆá‰Š-á‰á‰-á‰–á‰˜á‰š-á‰á‰ -áŠˆáŠŠ-áŠáŠ-áŠ°áŠ²-áŠµáŠ¸-áŠ¾á‹€á‹‚-á‹…á‹ˆ-á‹–á‹˜-áŒáŒ’-áŒ•áŒ˜-ášáŽ€-áŽáŽ -á´á-á™¬á™¯-á™¿áš-áššáš -á›ªá›®-á›°áœ€-áœŒáœŽ-áœ‘áœ -áœ±á€-á‘á -á¬á®-á°áž€-áž³áŸ—áŸœá  -á¡·á¢€-á¢¨á¢ªá¢°-á£µá¤€-á¤œá¥-á¥á¥°-á¥´á¦€-á¦«á§-á§‡á¨€-á¨–á¨ -á©”áª§á¬…-á¬³á…-á‹á®ƒ-á® á®®á®¯á®º-á¯¥á°€-á°£á±-á±á±š-á±½á³©-á³¬á³®-á³±á³µá³¶á´€-á¶¿á¸€-á¼•á¼˜-á¼á¼ -á½…á½ˆ-á½á½-á½—á½™á½›á½á½Ÿ-á½½á¾€-á¾´á¾¶-á¾¼á¾¾á¿‚-á¿„á¿†-á¿Œá¿-á¿“á¿–-á¿›á¿ -á¿¬á¿²-á¿´á¿¶-á¿¼â±â¿â‚-â‚œâ„‚â„‡â„Š-â„“â„•â„™-â„â„¤â„¦â„¨â„ª-â„â„¯-â„¹â„¼-â„¿â……-â…‰â…Žâ… -â†ˆâ°€-â°®â°°-â±žâ± -â³¤â³«-â³®â³²â³³â´€-â´¥â´§â´â´°-âµ§âµ¯â¶€-â¶–â¶ -â¶¦â¶¨-â¶®â¶°-â¶¶â¶¸-â¶¾â·€-â·†â·ˆ-â·Žâ·-â·–â·˜-â·žâ¸¯ã€…-ã€‡ã€¡-ã€©ã€±-ã€µã€¸-ã€¼ã-ã‚–ã‚-ã‚Ÿã‚¡-ãƒºãƒ¼-ãƒ¿ã„…-ã„ã„±-ã†Žã† -ã†ºã‡°-ã‡¿ã€-ä¶µä¸€-é¿Œê€€-ê’Œê“-ê“½ê”€-ê˜Œê˜-ê˜Ÿê˜ªê˜«ê™€-ê™®ê™¿-êš—êš -ê›¯êœ—-êœŸêœ¢-êžˆêž‹-êžŽêž-êž“êž -êžªêŸ¸-ê ê ƒ-ê …ê ‡-ê Šê Œ-ê ¢ê¡€-ê¡³ê¢‚-ê¢³ê£²-ê£·ê£»ê¤Š-ê¤¥ê¤°-ê¥†ê¥ -ê¥¼ê¦„-ê¦²ê§ê¨€-ê¨¨ê©€-ê©‚ê©„-ê©‹ê© -ê©¶ê©ºêª€-êª¯êª±êªµêª¶êª¹-êª½ê«€ê«‚ê«›-ê«ê« -ê«ªê«²-ê«´ê¬-ê¬†ê¬‰-ê¬Žê¬‘-ê¬–ê¬ -ê¬¦ê¬¨-ê¬®ê¯€-ê¯¢ê°€-íž£íž°-íŸ†íŸ‹-íŸ»ï¤€-ï©ï©°-ï«™ï¬€-ï¬†ï¬“-ï¬—ï¬ï¬Ÿ-ï¬¨ï¬ª-ï¬¶ï¬¸-ï¬¼ï¬¾ï€ïïƒï„ï†-ï®±ï¯“-ï´½ïµ-ï¶ï¶’-ï·‡ï·°-ï·»ï¹°-ï¹´ï¹¶-ï»¼ï¼¡-ï¼ºï½-ï½šï½¦-ï¾¾ï¿‚-ï¿‡ï¿Š-ï¿ï¿’-ï¿—ï¿š-ï¿œ]/: Range out of order in character class
    at new RegExp (<anonymous>)
    at Object.<anonymous> (keras.js:31478)
    at keras.js:31221
    at Object.e.read.s (keras.js:31221)
    at e (keras.js:1)
    at Object.<anonymous> (keras.js:31221)
    at Object.o (keras.js:31221)
    at e (keras.js:1)
    at Object.<anonymous> (keras.js:1)
    at e (keras.js:1)
    at Object.<anonymous> (keras.js:1)
    at e (keras.js:1)
    at Object.t.__esModule.default (keras.js:10644)
    at e (keras.js:1)
    at Object.t.__esModule.default (keras.js:10644)
    at e (keras.js:1)

Can you tell me why this happened and how to deal with it?

Best wishes, Yiming

opened by yiminglin-ai 3

Error when call model.predict

It's saying that there are a token |= but my code only has one file index.html and it has no such symbol. What is the cause here?

SyntaxError: Unexpected token |=
    at new Function (<anonymous>)
    at u (https://off99555.github.io/HoNNeT/keras-js/keras.js:13:8807)
    at assigns_cwise_thunk (eval at r (https://off99555.github.io/HoNNeT/keras-js/keras.js:13:10619), <anonymous>:7:29)
    at Object.assigns_ndarrayops [as assigns] (eval at i (https://off99555.github.io/HoNNeT/keras-js/keras.js:1:6744), <anonymous>:3:44)
    at e.value (https://off99555.github.io/HoNNeT/keras-js/keras.js:7:19739)
    at e.value (https://off99555.github.io/HoNNeT/keras-js/keras.js:7:21879)
    at t.value (https://off99555.github.io/HoNNeT/keras-js/keras.js:4:15902)
    at t.<anonymous> (https://off99555.github.io/HoNNeT/keras-js/keras.js:4:16786)
    at o (https://off99555.github.io/HoNNeT/keras-js/keras.js:16:7592)
    at Generator._invoke (https://off99555.github.io/HoNNeT/keras-js/keras.js:16:9001)
    at Generator.t.(anonymous function) [as next] (https://off99555.github.io/HoNNeT/keras-js/keras.js:16:7771)
    at Generator.i (https://off99555.github.io/HoNNeT/keras-js/keras.js:11:1532)
    at u._promiseFulfilled (https://off99555.github.io/HoNNeT/keras-js/keras.js:9:22799)
    at t.<anonymous> (https://off99555.github.io/HoNNeT/keras-js/keras.js:9:24615)
From previous event:
    at t (https://off99555.github.io/HoNNeT/keras-js/keras.js:4:15972)
    at t.<anonymous> (https://off99555.github.io/HoNNeT/keras-js/keras.js:4:18474)
    at o (https://off99555.github.io/HoNNeT/keras-js/keras.js:16:7592)
    at Generator._invoke (https://off99555.github.io/HoNNeT/keras-js/keras.js:16:9001)
    at Generator.t.(anonymous function) [as next] (https://off99555.github.io/HoNNeT/keras-js/keras.js:16:7771)
From previous event:
    at t (https://off99555.github.io/HoNNeT/keras-js/keras.js:4:17225)
    at model.ready.then (https://off99555.github.io/HoNNeT/:165:16)
From previous event:
    at https://off99555.github.io/HoNNeT/:158:20

Here's the javascript part of that index.html file:

     const N_AVAIL_HEROES = 260
     const HERO_INPUT = new Array(2 * N_AVAIL_HEROES).fill(0)
     const vm = new Vue({
         el: '#app',
         data() {
             return {
                 model: null,
                 modelReady: false
             }
         },
         watch: {
             modelReady: function() {
                 if (this.modelReady)
                     console.log('the model is ready now')
             }
         }
     })
     const model = new KerasJS.Model({
         filepaths: {
             model: 'honnet_brain_architecture.json',
             weights: 'honnet_brain_weights_weights.buf',
             metadata: 'honnet_brain_weights_metadata.json'
         }
     })
     function getInputData(legion, hellbourne) {
        console.log('predicting with data')
        const inputData = {
            'hero_input': new Float32Array(HERO_INPUT)
        }
        return inputData
     }
     model.ready().then(() => {
         vm.model = model
         vm.modelReady = true

         // make predictions
         // outputData is an object keyed by names of the output layers
         // or `output` for Sequential models
         model.predict(getInputData([], [])).then(function(outputData) { // <====== error happens here!
             console.log('output')
             console.log(outputData)
         }).catch(err => {
             // handle error
             console.log('pred error: ')
             console.log(err)
         })
     }).catch(err => {
         // handle error
         console.log('ready error: ')
         console.log(err)
     })

opened by off99555 3

[Model] path to protobuf-serialized model definition file is missing

Update from master 11/10/2017

Code builds but fails in browser, see error below. What changed?

Error: [Model] path to protobuf-serialized model definition file is missing. new Model ./src/Model.js:34 31 | } = config 32 | 33 | if (!filepath) {

34 | throw new Error('[Model] path to protobuf-serialized model definition file is missing.') 35 | } 36 | this.filepath = filepath 37 |

opened by dparnellmitek 2
Purpose of wrapping the prediction inference call in a setTimeout() and also passing in exactly 10 milliseconds

I have a small question, could anyone please explain to me what's the purpose of wrapping the prediction inference call in a setTimeout() and also passing in exactly 10 milliseconds (and not something like 5 or 15 milliseconds)? Thanks!

https://github.com/transcranial/keras-js/blob/7713c554f062a77b6897892c2d10209e72576868/demos/src/components/common/Imagenet.vue#L274-L276

opened by jackylu0124 0
Can i use CRF models on keras-js ?

Hello ! I have created a Named Entity Recognition model with tensorflow/keras/keras_contrib. I would like run this model on my clients browser with Keras.js but i doesn't found anything for this. When i load my model, i use keras_contrib class like this in python :

self.model = load_model( dir_path + '/v1/tools/TextAnalyzerData/model.h5', custom_objects={ 'CRF':CRF, 'crf_loss':crf_loss, 'crf_viterbi_accuracy':crf_viterbi_accuracy } )

Any alternative or tips for javascript size ? Thanks

opened by sandro971 0
model.predict() always outputs same invalid results after deployment

Problem Statment:

Goal: Use keras-js (https://github.com/transcranial/keras-js) to do Sentiment Analysis for input summary(NLP).

Developing environment: React 16.6.3

Related Dependencies: keras-js, imdb_bidirectional_lstm.bin(trained model, located in public folder), imdb_dataset_word_index_top20000.json

Problem: The model.predict() function in keras-js always output the same result, regardless what the the actual input is, the output is always 50% after deployment.

It works fine on my machine, I’m using Mac OS. But if we deployed it to Linux server, the problem occurred. We can load all dependencies in Linux server, no error messages, but the prediction results are totally different from that from local machine even we used completely same code. The prediction results in deployed version are always 50%, which is not valid.

opened by zjtisme 0

TypeError: layer.layer is undefined

I want to run a keras model on a browser using keras.js , but keras.js predict is not working. while loading i got the following error

TypeError: layer.layer is undefined here is the js code:

const model = new KerasJS.Model(
     {
     filepath: 'example.bin',
     gpu : true,
     filesystem : true

     })

       model.ready().then(() => model.predict(
       {
         input: new Float32Array(samples),
       })) .then(({output}) => 
       {
         alert(output['output']);
       })
       .catch((error) => 
       {
         alert(error);
      });

opened by hmhwe 0

Error: predict() must take an object where the keys are the named inputs of the model:

I have been trying to run a keras model in browser using keras.js. But whenever i try to predict I got the following error

Error: predict() must take an object where the keys are the named inputs of the model: Here is js code

handlePredict()
  {
    const model = new KerasJS.Model({
    filepaths: {
    model: 'model.json',
    weights: 'model_weights.buf',
    metadata: 'model_metadata.json'
  },

  gpu : true,
  filesystem : true

})

  model.ready().then(() =>{
    const inputData = {
      'input_1': new Float32Array(samples)
    }
    alert('Converted text = ' + inputData['input_1'])
    return model.predict(inputData['input_1'])
  })
  .then(output =>
    {
    alert('Predicted value : ' + output['fc1000']);
    })
  .catch(err =>
    {
    alert(err);
    })

  }

I am using keras-js 0.3.0 and keras 2.2.4 . Any help would be appreciated.

opened by hmhwe 0

Releases(v0.3.0)

v0.3.0(May 20, 2017)
Added back legacy Merge, MaxoutDense, Highway layers

Fixed model creation when Model class contains branches of Sequential class

Added SqueezeNet v1.1 and Auxiliary Classifier Generative Adversarial Network (AC-GAN) demos

Source code(tar.gz)
Source code(zip)
v0.2.0(May 20, 2017)

Updated library to be compatible with Keras v2 (tested on 2.0.4 with TF 1.1.0).
Source code(tar.gz)
Source code(zip)
v0.1.0(May 20, 2017)

Add as npm package for Node.js in addition to UMD browser build.
Source code(tar.gz)
Source code(zip)

Owner

Leon Chen

Co-founder, MD.ai. Software/ML engineer, doctor, electronic musician.

GitHub Repository https://transcranial.github.io/keras-js

Code for ACL2021 paper Consistency Regularization for Cross-Lingual Fine-Tuning.

xTune Code for ACL2021 paper Consistency Regularization for Cross-Lingual Fine-Tuning. Environment DockerFile: dancingsoul/pytorch:xTune Install the f

42 Dec 09, 2022

Pervasive Attention: 2D Convolutional Networks for Sequence-to-Sequence Prediction

This is a fork of Fairseq(-py) with implementations of the following models: Pervasive Attention - 2D Convolutional Neural Networks for Sequence-to-Se

490 Dec 15, 2022

A PaddlePaddle version image model zoo.

Paddle-Image-Models English | 简体中文 A PaddlePaddle version image model zoo. Install Package Install by pip： $ pip install ppim Install by wheel package

131 Dec 07, 2022

SAS: Self-Augmentation Strategy for Language Model Pre-training

SAS: Self-Augmentation Strategy for Language Model Pre-training This repository

5 Nov 02, 2022

Efficient and Scalable Physics-Informed Deep Learning and Scientific Machine Learning on top of Tensorflow for multi-worker distributed computing

Notice: Support for Python 3.6 will be dropped in v.0.2.1, please plan accordingly! Efficient and Scalable Physics-Informed Deep Learning Collocation-

74 Dec 09, 2022

HiFi++: a Unified Framework for Neural Vocoding, Bandwidth Extension and Speech Enhancement

HiFi++ : a Unified Framework for Neural Vocoding, Bandwidth Extension and Speech Enhancement This is the unofficial implementation of Vocoder part of

118 Dec 29, 2022

Machine Learning Models were applied to predict the mass of the brain based on gender, age ranges, and head size.

Brain Weight in Humans Variations of head sizes and brain weights in humans Kaggle dataset obtained from this link by Anubhab Swain. Image obtained fr

1 Feb 02, 2022

Line-level Handwritten Text Recognition (HTR) system implemented with TensorFlow.

Line-level Handwritten Text Recognition with TensorFlow This model is an extended version of the Simple HTR system implemented by @Harald Scheidl and

72 May 07, 2022

A library of scripts that interact with the PythonTurtle module to create games, drawings, and more

TurtleLib TurtleLib is a library of scripts that interact with the PythonTurtle module to create games, drawings, and more! Using the Scripts Copy or

1 Jan 15, 2022

Tiny-NewsRec: Efﬁcient and Effective PLM-based News Recommendation

Tiny-NewsRec The source codes for our paper "Tiny-NewsRec: Efﬁcient and Effective PLM-based News Recommendation". Requirements PyTorch == 1.6.0 Tensor

3 Dec 07, 2022

[CVPR 2022] Official code for the paper: "A Stitch in Time Saves Nine: A Train-Time Regularizing Loss for Improved Neural Network Calibration"

MDCA Calibration This is the official PyTorch implementation for the paper: "A Stitch in Time Saves Nine: A Train-Time Regularizing Loss for Improved

21 Dec 22, 2022

Run Keras models in the browser, with GPU support using WebGL

Related tags

Overview

Interactive Demos | Documentation

Comments

Releases(v0.3.0)

v0.3.0(May 20, 2017)

v0.2.0(May 20, 2017)

v0.1.0(May 20, 2017)

Owner

Leon Chen

Code for ACL2021 paper Consistency Regularization for Cross-Lingual Fine-Tuning.

Pervasive Attention: 2D Convolutional Networks for Sequence-to-Sequence Prediction

A PaddlePaddle version image model zoo.

SAS: Self-Augmentation Strategy for Language Model Pre-training

Efficient and Scalable Physics-Informed Deep Learning and Scientific Machine Learning on top of Tensorflow for multi-worker distributed computing

HiFi++: a Unified Framework for Neural Vocoding, Bandwidth Extension and Speech Enhancement

Machine Learning Models were applied to predict the mass of the brain based on gender, age ranges, and head size.

Line-level Handwritten Text Recognition (HTR) system implemented with TensorFlow.

A library of scripts that interact with the PythonTurtle module to create games, drawings, and more

Tiny-NewsRec: Efﬁcient and Effective PLM-based News Recommendation

[CVPR 2022] Official code for the paper: "A Stitch in Time Saves Nine: A Train-Time Regularizing Loss for Improved Neural Network Calibration"

An educational resource to help anyone learn deep reinforcement learning.

Hierarchical Motion Encoder-Decoder Network for Trajectory Forecasting (HMNet)

A Robust Unsupervised Ensemble of Feature-Based Explanations using Restricted Boltzmann Machines

[CVPR 2022 Oral] Crafting Better Contrastive Views for Siamese Representation Learning

Official implementation of NeurIPS'2021 paper TransformerFusion

This repository contains the official code of the paper Equivariant Subgraph Aggregation Networks (ICLR 2022)

A fast MoE impl for PyTorch

Show-attend-and-tell - TensorFlow Implementation of "Show, Attend and Tell"

PyTorch implementation of the paper The Lottery Ticket Hypothesis for Object Recognition